Intelligent video software (IVS), or video analytics, is a real revolution in the digital surveillance world. While DVRs and network cameras make people watch more, IVS allows people to watch no more. IVS makes it possible to advance crisis management from after the event to during the event, and even before the event (loitering detection, for example). There is no doubt IVS has given surveillance a whole new meaning.
When talking about IVS or video analytics, it is an open, undefined topic. For example, when people counting is mentioned, some think about megastores while others think about outdoor New Year parties.
Currently, the most widely adopted function is intrusion detection, including per imeter fencing in its broadest sense. Especially for wide -area surveillance such as airports, coastlines, borders or courtyards of mansions, intrusion detection is a fairly mature solution, with a comparatively lower cost. Other available functions include l i cense plate recogni t ion, face detection and object counting.
Take people counting. If multiple cameras are installed at different locations in a megastore, statistical analysis can be done to improve store sales or customer experience.
Vehicle counting is another application in high demand. Just like people counting, density affects the accuracy of vehicle counting as well. There should still be some distance between two cars, no matter how congested traffic is. However, an IVS system cannot detect that unless cameras are mounted to look downward vertically.
Facial recognition, which is different from face detection, is also on the rise. But even with the most advanced technology available today, it is still difficult to pick out a specific person from a database of thousands of people, let alone the computation cost if the processing is done in real time.
Generally speaking, a video analytic function is composed of multiple processing procedures. The task for each procedure is to extract useful bits from the previous procedure, and then pass them to the next procedure. The process repeats itself over and over until it is possible to determine ¨yes〃 or ¨no〃 or, in the case of object counting, the actual number. Quality of a video analytic function, thus, depends on how well defined each procedure is, what procedures are involved and what their order is, and how finely tuned the parameters are.
As far as practicality is concerned, another important factor other than parameters and algor ithms that cannot be overlooked is cost, which must be acceptable by the market. There is always a tradeoff between computation power and accuracy. Fortunately, the cost of computation has been declining for the past few years.
The Near Future
In the early days, video analytics usually appeared in the form of pure software with video capture support. Today, video analytics can be found embedded in stand-alone DVRs or network cameras. While intelligence is moving out to front-end edge devices, back-end software is still critical in overall operation:
1. In a system with multiple network cameras , communi cat ion and int egr a t ion among di f fe rent cameras still rely on back-end software. Even though front-end devices help save computation and bandwidth costs, inter-camera analysis and the cross examination between video and non-video information would still have to be done by the back-end software.
2. No matter how smart edge devices get, the information they gather will still have to be delivered to the back end eventual ly. That is to say, front-end devices cannot be totally independent, at least in the near future.
3. Distributed processing always costs more than the centralized approach, not to mention copies of royalties needed.
4. Just because an IVS system does not detect abnormality does not mean everything is fine and dandy. In most cases, complete recordings are still needed for post-event analysis and prosecution.
Ultimately, accuracy and total cost of ownership are the deciding factors.
In the Long Run
Leading IVS players are all saying how important it is to communicate with users during the sales process. As communication is the best way to reduce gaps between user expectations and reality, IVS should not make its way to store shelves anytime soon. It is impossible for any given analytic function to work accurately in all scenarios. It is, however, possible to customize a set of parameters to typical scenarios in a specific vertical market.
Another issue that should not be neglected is the degree of integration for the whole system. As discussed above, the smarter the devices get, the greater the needs there will be for an even smarter management system to fully unleash the power of individual devices. No matter how far human eyes can see or what sounds human ears can hear, it is still the human brain that makes sense of everything.